Global cluster config not sent to memcached after memcached is OOM-killed and restarted

SDK tries to get configuration via KV connection, before selecting bucket, but the server returns NOT_FOUND (see tcp dump in the attachment).

This behaviour is reproducible with cluster "10.1.28.117:31995" only. Another cluster that uses same version works. Also this particular cluster reported as operational several days ago.

From looking at the logs, memcahced was restarted on 10th January due to the Linux OOM-killer terminating it. After restart, no global config is sent to memcached (per-bucket config is correctly sent):

$ rg -N '(Couchbase version|Updated global cluster configuration)' memcached.log 

2023-01-10T14:39:51.635909+00:00 INFO Couchbase version 7.1.2-3454 starting.

Note the lack of "Updated global cluster configuration" message, which is seen during initial CB Server startup (i.e. when babysitter and ns_server starts) - example from a different 7.1.1 system:

# grep -E '(Couchbase version|Updated global cluster configuration)' memcached.log.000084.txt 

2023-01-13T17:12:33.387179+00:00 INFO Couchbase version 7.1.1-0000 starting.

2023-01-13T17:12:34.112813+00:00 INFO 62: [ {"ip":"127.0.0.1","port":47476} - {"ip":"127.0.0.1","port":11209} (System, <ud>@ns_server</ud>) ] Updated global cluster configuration. New revision: {"epoch":1,"revno":4095}